Comparative analysis of machine learning classifiers and deep learning models for categorization of Knee Osteoarthritis

Knee Osteoarthritis also known as KOA is a disease in which there is so much pain, stiffness and it also limits the mobility of patient, if not cured at right time it can lead to disablement due to degeneration of articular cartilage in knee joint. Due to limited mobility , its treatment and diagnosis are very challenging, especially when there is lack of devices and technologies for precise identification and tracking of this disease’s progression at right time. There is a very common method known as “Kellgren-Lawrence (KL) grading” by which degree of Osteoarthritis is determined. The scale of KL grading ranges from ‘0’ to ‘4’ where 0 is ‘no osteoarthritis’ and 4 is ‘severe osteoarthritis. By using machine learning and deep learning, this work presented a approach that improves the accuracy of classification of KOA and its level diagnosis by using X-ray images. In this research work, feature extraction techniques such Global Average Pooling, Min–Max scaling, Histogram of Oriented Gradients (HOG) along with another technique called Linear Discriminant Analysis (LDA) applied on Xray dataset. The study evaluates two Machine Learning classifiers which were Support Vector Machine (SVM) and XGBoost which both are optimized through GridSearchCV for hyperparameter tuning and two deep learning models EfficientNetB6 and EfficientNetB7 which both were fine tuned. The proposed approach evaluates the knee X-ray images and assigns them to one of 0, 1, 2, 3, or 4 grades in order to automate KL grading. From the experimental results, it is concluded that the XGBoost classifier performed the best with 97.00 % accuracy.